Root initiated routing state in RPL
Cisco Systems
Village d'Entreprises Green Side400, Avenue de RoumanilleBatiment T3Biot - Sophia Antipolis06410FRANCE+33 4 97 23 26 34pthubert@cisco.comHuawei TechKundalahalli Village, Whitefield,BangaloreKarnataka560037India+91-080-49160700rahul.ietf@gmail.com
Cisco Systems
Cessna Business ParkKadubeesanahalliMarathalli ORRBangaloreKarnataka560087INDIA+91 80 4426 4140mundenma@cisco.com
Routing
ROLL
This document proposes a protocol extension to RPL that enables to
install a limited amount of centrally-computed routes in a RPL graph,
enabling loose source routing down a non-storing mode DODAG, or
transversal routes inside the DODAG.
As opposed to the classical route injection in RPL that are injected
by the end devices, this draft enables the root of the DODAG to
projects the routes that are needed on the nodes where they should be
installed.
The
"Routing Protocol for Low Power and Lossy Networks" (LLN)(RPL)
is a generic Distance Vector protocol that is well suited for application
in a variety of low energy Internet of Things (IoT) networks.
RPL forms Destination Oriented Directed Acyclic Graphs (DODAGs) in which
the root often acts as the Border Router to connect the RPL domain to the
Internet. The root is responsible to select the RPL Instance that is used
to forward a packet coming from the Internet into the RPL domain and set
the related RPL information in the packets.
The
6TiSCH architecture leverages RPL for its routing operation and
considers the
Deterministic Networking Architecture as one possible model
whereby the device resources and capabilities are exposed to an external
controller which installs routing states into the network based on some
objective functions that reside in that external entity.
Based on heuristics of usage, path length, and knowledge of device capacity
and available resources such as battery levels and reservable buffers, a
Path Computation Element () with a global visibility
on the system could install additional P2P routes that are more optimized
for the current needs as expressed by the objective function.
This draft enables a RPL root to install and maintain projected routes
(P-routes) within its DODAG, along a selected set of nodes that may or may
not include self, for a chosen duration. This potentially enables routes
that are more optimized than those obtained with the distributed operation
of RPL, either in terms of the size of a source-route header or in terms of
path length, which impacts both the latency and the packet delivery ratio.
P-routes may be installed in either Storing and Non-Storing Modes Instances
of the classical RPL operation, resulting in potentially hybrid situations
where the mode of some P-routes is different from that of the other routes
in the RPL Instance.
Projected routes must be used with the parsimony to limit the amount of
state that is installed in each device to fit within its resources, and to
limit the amount of rerouted traffic to fit within the capabilities of the
transmission links.
The algorithm used to compute the paths and the protocol used to learn the
topology of the network and the resources that are available in devices and
in the network are out of scope for this document.
Possibly with the assistance of a Path Computation Element
() that could have a better visibility on the
larger system, the root computes which segment could be optimized and uses
this draft to install the corresponding projected routes.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described in BCP 14
when, and only when, they
appear in all capitals, as shown here.
In this document, readers will encounter terms and concepts
that are discussed in the following documents:
"Routing Protocol for Low Power and Lossy Networks"
, and "Terminology in Low power And Lossy Networks"
. This document often uses the following acronyms:
6LoWPAN Backbone Router 6LoWPAN Border Router 6LoWPAN Node 6LoWPAN Router Capability Indication Option (Extended) Address Registration Option -- (E)ARO (Extended) Duplicate Address Request -- (E)DAR (Extended) Duplicate Address Confirmation --
(E)DAC Duplicate Address Detection Destination-Oriented Directed Acyclic Graph Low-Power and Lossy Network Neighbor Advertisement Neighbor Cache Entry Neighbor Discovery Neighbor Discovery Protocol Neighbor Solicitation IPv6 Routing Protocol for LLNs
(pronounced ripple) Router Advertisement Router Solicitation
A route that is installed remotely by a RPL root.
Section 6.7 of RPL specifies Control Message Options (CMO)
to be placed in RPL messages such as the Destination Advertisement Object
(DAO) message. The RPL Target Option
and the Transit Information Option (TIO) are such options; the former
indicates a node to be reached and the latter specifies
a parent that can be used to reach that node. Options may be factorized;
one or more contiguous TIOs apply to the one or more contiguous Target
options that immediately precede the TIOs in the RPL message.
This specification introduces 2 new Control Message Options referred to as
Route Projection Options (RPO). One RPO is the Information option (VIO) and
the other is the Source-Routed VIO (SRVIO). The VIO installs a route on each
hop along a projected route (in a fashion analogous to RPL Storing Mode)
whereas the SRVIO installs a source-routing state at the ingress node, which
uses it to insert a routing header in a fashion similar to Non-Storing Mode.
Like the TIO, the RPOs MUST be preceded by one or more RPL Target Options to
which they apply, and they can be factorized: multiple contiguous RPOs
indicate alternate paths to the target(s).
The format of RPOs is as follows:
0x0A for VIO, 0x0B for SRVIO
(to be confirmed by IANA)In bytes; variable, depending on the number
of Via Addresses.8-bit unsigned integer. When a RPL
Target option is issued by the root of the DODAG
(i.e. in a DAO message), that root sets the Path Sequence and
increments the Path Sequence each time it issues a RPL Target
option with updated information. The indicated sequence
deprecates any state for a given Target that was learned from a
previous sequence and adds to any state that was learned for
that sequence.8-bit unsigned integer. The length
of time in Lifetime Units (obtained from the Configuration
option) that the prefix is valid for route determination. The
period starts when a new Path Sequence is seen. A value of all
one bits (0xFF) represents infinity. A value of all zero bits
(0x00) indicates a loss of reachability. A DAO message that
contains a Via Information option with a Path Lifetime of
0x00 for a Target is referred as a No-Path (for that Target) in
this document.16 bytes. IPv6 Address of the
next hop towards the destination(s) indicated in the target option
that immediately precede the RPO. Via Addresses are indicated in
the order of the data path from the ingress to the egress nodes.
TBD: See how the /64 prefix can be elided if it is the same as
that of (all of) the target(s). In that case, the Next-Hop Address
could be expressed as the 8-bytes suffix only.
An RPO MUST contain at least one Via Address, and a Via Address MUST NOT be
present more than once, otherwise the RPO MUST be ignored.
This draft adds a capability to RPL whereby the root of a DODAG projects a
route by sending an extended DAO message called a Projected-DAO (P-DAO) to an
arbitrary router in the DODAG, indicating one or more sequence(s) of routers
inside the DODAG via which the target(s) indicated in the Target Information
Option(s) (TIO) can be reached.
A P-DAO is sent from a global address of the root to a global address of the
recipient, and MUST be confirmed by a DAO-ACK, which is sent back to a
global address of the root.
A P-DAO message MUST contain at least one TIO and at least one RPO following
it. There can be at most one such sequence of TIOs and then RPOs.
Like a classical DAO message, a P-DAO is processed only if it is "new"
per section 9.2.2. "Generation of DAO Messages" of the
RPL specification; this is determined using the Path Sequence
information from the RPO as opposed to a TIO. Also, a Path Lifetime of 0 in
an RPO indicates that a route is to be removed.
There are two kinds of operation for the projected routes, the Storing Mode
and the Non-Storing Mode.
The Non-Storing Mode is discussed in section .
It uses an SRVIO that carries a list of Via Addresses to be used as a
source-routed path to the target. The recipient of the P-DAO is the
ingress router of the source-routed path. Upon a Non-Storing Mode P-DAO, the
ingress router installs a source-routed state to the target and replies to
the root directly with a DAO-ACK message.
The Storing Mode is discussed in section . It uses
a VIO with one Via Address per consecutive hop, from the ingress to the
egress of the path, including the list of all intermediate routers in the
data path order.
The Via Addresses indicate the routers in which the routing state to the
target have to be installed via the next Via Address in the VIO.
In normal operations, the P-DAO is propagated along the chain of Via Routers
from the egress router of the path till the ingress one, which confirms the
installation to the root with a DAO-ACK message.
Note that the root may be the ingress and it may be the egress of the
path, that it can also be neither but it cannot be both.
As illustrated in , a P-DAO that carries an SRVIO
enables the root to install a source-routed path towards a target in any
particular router; with this path information the router can add a source
routed header reflecting the P-route to any packet for which the current
destination either is the said target or can be reached via the target.
A route indicated by an SRVIO may be loose, meaning that the node that owns
the next listed Via Address is not necessarily a neighbor. Without proper
loop avoidance mechanisms, the interaction of loose source routing and other
mechanisms may effectively cause loops. In order to avoid those loops, if
the router that installs a P-route does not have a connected route
(a direct adjacency) to the next soure routed hop and fails to locate it as
a neighbor or a neighbor of a neighbor, then it MUST ensure that it has
another projected route to the next loose hop under the control of the same
route computation system, otherwise the P-DAO is rejected.
When forwarding a packet to a destination for which the router determines
that routing happens via the target, the router inserts the source routing
header in the packet to reach the target. In the case of a loose
source-routed path, there MUST be either a neighbor that is adjacent to the
loose next hop, on which case the packet s forwarded to that neighbor, or
a source-routed path to the loose next hop; in the latter case,
another encapsulation takes place and the process possibly recurses;
otherwise the packet is dropped.
In order to add a source-routing header, the router encapsulates the
packet with an IP-in-IP header and a non-storing mode source routing
header (SRH) .
In the uncompressed form the source of the packet would be self, the
destination would be the first Via Address in the SRVIO, and the SRH would
contain the list of the remaining Via Addresses and then the target.
In practice, the router will normally use the
"IPv6 over Low-Power Wireless Personal Area Network (6LoWPAN) Paging
Dispatch" to compress the RPL artifacts as indicated in the
"6LoWPAN Routing Header"
specification. In that case, the router indicates self as
encapsulator in an IP-in-IP 6LoRH Header, and places the list of Via
Addresses in the order of the VIO and then the target in the SRH 6LoRH
Header.
As illustrated in , the Storing Mode projected iq
used by the root to install a routing state towards a target in the routers
along a segment between an ingress and an egress router;
this enables the routers to forward along that segment any packet for
which the next loose hop is the said target, for instance a loose source
routed packet for which the next loose hop is the target, or a packet for
which the router has a routing state to the final destination via the target.
In order to install the relevant routing state along the segment between an
ingress and an egress routers,
the root sends a unicast P-DAO message to the egress router of the routing
segment that must be installed. The P-DAO message contains the ordered list
of hops along the segment as a direct sequence of Via Information options
that are preceded by one or more RPL Target options to which they relate.
Each Via Information option contains a Path Lifetime for which the state is
to be maintained.
The root sends the P-DAO directly to the egress node of the segment.
In that P-DAO, the destination IP address matches the Via Address in the
last VIO. This is how the egress recognizes its role. In a similar fashion,
the ingress node recognizes its role as it matches Via Address in the first
VIO.
The egress node of the segment is the only node in the path that does not
install a route in response to the P-DAO; it is expected to be already able
to route to the target(s) on its own. It may either be the target, or may
have some existing information to reach the target(s), such as a connected
route or an already installed projected route.
If one of the targets cannot be located, the node MUST answer to the root
with a negative DAO-ACK listing the target(s) that could not be located
(suggested status 10 to be confirmed by IANA).
If the egress node can reach all the targets, then it forwards the P-DAO
with unchanged content to its loose predecessor in the segment as indicated
in the list of Via Information options, and recursively the message is propagated
unchanged along the sequence of routers indicated in the P-DAO, but in the
reverse order, from egress to ingress.
The address of the predecessor to be used as destination of the propagated
DAO message is found in the Via Information option the precedes the one
that contain the address of the propagating node, which is used as source
of the packet.
Upon receiving a propagated DAO, an intermediate router as well as the
ingress router install a route towards the DAO target(s) via its
successor in the P-DAO; the router locates the VIO that contains its
address, and uses as next hop the address found in the Via Address field
in the following VIO. The router MAY install additional routes towards the
addresses that are located in VIOs that are after the next one, if any, but
in case of a conflict or a lack of resource, a route to a target installed
by the root has precedence.
The process recurses till the P-DAO is propagated to ingress router of
the segment, which answers with a DAO-ACK to the root.
Also, the path indicated in a P-DAO may be loose, in which case the
reachability to the next hop has to be asserted. Each router along the
path indicated in a P-DAO is expected to be able to reach its successor,
either with a connected route (direct neighbor), or by routing, for instance
following a route installed previously by a DAO or a P-DAO message.
If that route is not connected then a recursive lookup may take place at
packet forwarding time to find the next hop to reach the target(s).
If it does not and cannot reach the next router in the P-DAO,
the router MUST answer to the root with a negative DAO-ACK
indicating the successor that is unreachable
(suggested status 11 to be confirmed by IANA).
A Path Lifetime of 0 in a Via Information option is used to clean up the
state. The P-DAO is forwarded as described above, but the DAO
is interpreted as a No-Path DAO and results in cleaning up existing state
as opposed to refreshing an existing one or installing a new one.
A RPL implementation operating in a very constrained LLN typically uses
the Non-Storing Mode of Operation as represented in .
In that mode, a RPL node indicates a
parent-child relationship to the root, using a Destination Advertisement
Object (DAO) that is unicast from the node directly to the root,
and the root typically builds a source routed path to a destination down
the DODAG by recursively concatenating this information.
Based on the parent-children relationships expressed in the non-storing
DAO messages,the root possesses topological information about the whole
network, though this information is limited to the structure of the DODAG
for which it is the destination.
A packet that is generated within the domain will always reach the root,
which can then apply a source routing information to reach the destination
if the destination is also in the DODAG.
Similarly, a packet coming from the outside of the domain for a destination
that is expected to be in a RPL domain reaches the root.
It results that the root, or then some associated centralized computation
engine such as a PCE, can determine the amount of packets that reach a
destination in the
RPL domain, and thus the amount of energy and bandwidth that is wasted for
transmission, between itself and the destination, as well as the risk of
fragmentation, any potential delays because of a paths longer than
necessary (shorter paths exist that would not traverse the root).
As a network gets deep, the size of the source routing header that the
root must add to all the downward packets becomes an issue for nodes that
are many hops away. In some use cases, a RPL network forms long lines and
a limited amount of well-targeted routing state would allow to make the
source routing operation loose as opposed to strict, and save packet size.
Limiting the packet size is directly beneficial to the energy budget, but,
mostly, it reduces the chances of frame loss and/or packet fragmentation,
which is highly detrimental to the LLN operation. Because the capability
to store a routing state in every node is limited, the decision of which
route is installed where can only be optimized with a global knowledge of
the system, a knowledge that the root or an associated PCE may possess by
means that are outside of the scope of this specification.
This specification enables to store source-routed or storing mode state in
intermediate routers, which enables to limit the excursion of the source
route headers in deep networks.
Once a P-DAO exchange has taken place for a given target, if the root
operates in non storing mode, then it may elide the sequence of routers
that is installed in the network from its source route headers to
destination that are reachable via that target, and the source route
headers effectively become loose.
RPL is optimized for Point-to-Multipoint
(P2MP), root to leaves and Multipoint-to-Point (MP2P) leaves to root operations,
whereby routes are always installed along the RPL DODAG. Transversal
Peer to Peer (P2P) routes in a RPL network will generally suffer from some
stretch since routing between 2 peers always happens via a common parent,
as illustrated in :
in non-storing mode, all packets
routed within the DODAG flow all the way up to the root of the DODAG. If
the destination is in the same DODAG, the root must encapsulate the packet
to place a Routing Header that has the strict source route information down
the DODAG to the destination. This will be the case even if the destination
is relatively close to the source and the root is relatively far off.
In storing mode, unless the destination is a child of the source,
the packets will follow the default route up the DODAG as well.
If the destination is in the same DODAG, they will eventually reach a
common parent that has a route to the destination; at worse, the common
parent may also be the root. From that common parent, the packet will
follow a path down the DODAG that is optimized for the Objective Function
that was used to build the DODAG.
It results that it is often beneficial to enable transversal P2P routes,
either if the RPL route presents a stretch from shortest path, or if the
new route is engineered with a different objective.
For that reason, earlier work at the IETF introduced the
"Reactive Discovery of Point-to-Point Routes in
Low Power and Lossy Networks", which specifies a distributed method for
establishing optimized P2P routes. This draft proposes an alternate based
on a centralized route computation.
This specification enables to store source-routed or storing mode state in
intermediate routers, which enables to limit the stretch of a P2P route
and maintain the characteristics within a given SLA. An example of service
using this mechanism oculd be a control loop that would be installed in a
network that uses classical RPL for asynchronous data collection. In that
case, the P2P path may be installed in a different RPL Instance, with a
different objective function.
It must be noted that RPL has a concept of instance but does not have a
concept of an administrative distance, which exists in certain proprietary
implementations to sort out conflicts between multiple sources of routing
information. This draft conforms the instance model as follows:
If the PCE needs to influence a particular instance to add better routes
in conformance with the routing objectives in that instance, it may do so.
When the PCE modifies an existing instance then the added routes
must not create a loop in that instance. This is achieved by always
preferring a route obtained from the PCE over a route that is learned via
RPL.
If the PCE installs a more specific (say, Traffic Engineered) route between
a particular pair of nodes then it SHOULD use a Local Instance from the
ingress node of that path. A packet associated with that instance will
be routed along that path and MUST NOT be placed over a Global Instance
again. A packet that is placed on a Global Instance may be injected in the
Local Instance based on node policy and the Local Instance paramenters.
In all cases, the path is indicated by a new Via Information option, and
the flow is similar to the flow used to obtain loose source routing.
This draft uses messages that are already present in RPL
with optional secured versions. The same secured
versions may be used with this draft, and whatever security is deployed for
a given network also applies to the flows in this draft.
This document extends the IANA registry created by RFC 6550 for RPL
Control Codes as follows:CodeDescriptionReference0x0AViaThis document0x0BSource-Routed ViaThis documentThis document is updating the registry created by RFC 6550 for the RPL
3-bit Mode of Operation (MOP) as follows:
MOP valueDescriptionReference5Non-Storing mode of operation with Projected routesThis document6Storing mode of operation with Projected routesThis documentThe authors wish to acknowledge JP Vasseur and Patrick Wetterwald for their
contributions to the ideas developed here.Path Computation ElementIETF
In non-storing mode, the DAG root maintains the knowledge of the whole DODAG
topology, so when both the source and the destination
of a packet are in the DODAG, the root can determine the common
parent that would have been used in storing mode, and thus the list of nodes
in the path between the common parent and the destination. For instance in
the diagram shown in , if the source is node 41
and the destination is node 52, then the common parent is node 22.
With this draft, the root can install a storing mode routing states along a
segment that is either from itself to the destination, or from one or more
common parents for a particular source/destination pair towards that
destination (in this particular
example, this would be the segment made of nodes 22, 32, 42).
In the example below, say that there is a lot of traffic to nodes 55 and
56 and the root decides to reduce the size of routing headers to those
destinations. The root can first send a DAO to node 45 indicating target 55
and a Via segment (35, 45), as well as another DAO to node 46 indicating
target 56 and a Via segment (35, 46). This will save one entry in the
routing header on both sides. The root may then send a DAO to node 35
indicating targets 55 and 56 a Via segment (13, 24, 35) to fully optimize
that path.
Alternatively, the root may send a DAO to node 45 indicating target 55
and a Via segment (13, 24, 35, 45) and then a DAO to node 46 indicating
target 56 and a Via segment (13, 24, 35, 46), indicating the same DAO
Sequence.
In this example, say that a PCE determines that a path must be installed
between node S and node D via routers A, B and C, in order to serve the needs
of a particular application.
The root sends a P-DAO with a target option indicating the destination D and
a sequence Via Information option, one for S, which is the ingress router of
the segment, one for A and then for B, which are an intermediate routers, and
one for C, which is the egress router.
Upon reception of the P-DAO, C validates that it can reach D, e.g. using
IPv6 Neighbor Discovery, and if so, propagates the P-DAO unchanged to B.
B checks that it can reach C and of so, installs a route towards D via C.
Then it propagates the P-DAO to A.
The process recurses till the P-DAO reaches S, the ingress of the segment,
which installs a route to D via A and sends a DAO-ACK to the root.
As a result, a transversal route is installed that does not need to follow
the DODAG structure.